Validation of Visually Interpreted Corine Land Cover Classes with Spectral Values of Satellite Images and Machine Learning

We analyzed the Corine Land Cover 2018 (CLC2018) dataset to reveal correspondence between land cover categories of CLC and spectral information Landsat-8, Sentinel-2 PlanetScope images. Level 1 CLC2018 were in a 25 km × study area Hungary. Spectral data summarized by polygons, was evaluated with statistical tests. then performed Linear Discriminant Analysis (LDA) Random Forest classifications if L1 level confirmed values. Wetlands water bodies most likely be confused other categories. The least mixture observed when we applied median quantify pixel variance polygons. RF outperformed LDA’s accuracy, PlanetScope’s accurate. class accuracies showed that agricultural areas wetlands had issues misclassification. proved representativeness results repeated randomized test, only seemed ungeneralizable. Results as basic units cover, can ensure 71.1–78.5% OAs for three satellite sensors; higher geometric resolution resulted better accuracy. These justified spite visual interpretation, hold relevant about considering surface reflectance values satellites. However, using ground truth questionable, at nomenclature.